In this project, we intended to reverse the traditional workflow and use the historical maps to validate remote-sensing based assessment of temporal changes in development over our regional-scale study area. We also intended to use the map information as a proxy for development to compliment the remote sensing data. Image classification partitions image pixels or segments into unique categories based on spectral characteristics. Very narrow features, such as roads, are often neglected in classification procedures. The resulting classification tends to under represent the total area of developed land or impervious surface. It was our intension to merge information on roadways from historical maps with remote sensing data to improve classification of developed land. However, we were confronted with numerous challenges in this approach.

We discovered a source for historical maps at the University of Alabama Cartographic Research Laboratory, Tuscaloosa, Alabama. Their entire inventory of historical county-based paper maps was in digital form, having been scanned in a preservation effort. The information content of these maps comprises interstates, U.S., state, and local roadways within incorporated and unincorporated areas. A variety of other features are noted including water sources, railroads, airports, cultural/institutional landmarks, and a variety of other miscellaneous characteristics are also noted.

• A complete listing of the transportation maps available can be found here.
• A graphical view of historical map coverage for different time periods can be found here.
• A representative pair of transportation map subsets showing typical levels of change can be found here.

Map Registration
The digital maps were not registered to a particular earth grid for use in a GIS. Ultimately, all maps needed to be registered to a common grid used for the remote sensing data as well. It was noted that due to the poor quality and/or the scans thereof some maps were missing legible fiducial marks to use for registration.

Temporal Coverage
The scanned transportation maps span time from the early 1970's to 1999. Data availability by county ranged from one to four varying time periods. Each county updated its map as resources permitted. In some cases, maps were updated every 5-10 years, but in quite a few counties, we were only able to acquire a single historical map. The remote sensing classifications were nominally at 1980, 1990, and 2000. Map dates did not necessarily coincide with these dates, which left numerous uncertainties as to how to address such temporal discontinuities.

Data Quality
The historical maps that were scanned were by their very nature aged. Presumably, each county loaned a set of historical maps to the University of Alabama for scanning and archiving. It is readily apparent that the quality of some of the maps was poor resulting in a poor archive product. The general quality of historical data is often not up to current standards. Consistency and/or continuity of accuracy standards may not be maintained through time requiring different standards for different data sources from different periods. Metadata may not be available for the earlier data sets making it difficult to assess data quality and define accuracy retroactively. Map quality plays a significant role in the ability of linear feature extraction algorithms to perform without a significant number of breaks, which must be treated manually by the image analyst.

Linear Feature Extraction Algorithm
A linear feature extraction algorithm was to be applied to each of the digital images in order to extract linear features (roads and boundary lines). The process is described in more detail here. After the road network on each map was vectorized, it could be attributed as to date, number of lanes, etc. However, due to image quality issues, the need to "train" the algorithm and continuously monitor and intervene in the process, it proved to be impractical to continue this process for the entire set of maps. Uncertainties about the product given the issues noted herein did not warrant further labor investment.